稀疏特征重用的人脸特征提取网络

为了提高人脸特征提取网络的性能,进而提高人脸识别算法的准确率,本文对基于卷积神经网络的人脸特征提取网络进行研究,提出了SFRNet (Sparse Feature Reuse Network)。首先,基于稀疏特征重用、混合特征融合、中心-高斯池化三个创新点,给出了SFRNet的网络结构。然后,在图像分类数据集ImageNet和人脸识别数据集LFW (Labeled Faces in the Wild)、MegaFace上进行实验,分别验证了SFRNet在一般场景和人脸识别这一特定场景下的特征提取能力。实验表明本文所设计的SFRNet不仅计算量和参数量小,还能有效提取到人脸特征并且在一般场景中也有较强的泛化能力。 

Face Feature Extraction Network Based on Sparse Feature Reuse

In order to improve the performance of the face feature extraction network and then the accuracy of the face recognition algorithm, this paper studies the network of the face feature extractor based on the convolutional neural network and proposes SFRNet (Sparse Feature Reuse Network). First, based on the three innovations of sparse feature reuse, hybrid feature fusion, and Center-Gaussian pooling, the network structure of SFRNet is given. Then, experiments were performed on the image classification dataset ImageNet, the face recognition dataset LFW (Labeled Faces in the Wild), and MegaFace, respectively, to verify the feature extraction capabilities of SFRNet in the general scene and the specific scene of face recognition. Experiments show that the SFRNet designed in this article not only has a small amount of calculation and parameters, but also can effectively extract facial features and has strong generalization ability in general scenes.